Segmentation Aware Probabilistic Phenotyping Of Singlecell Spatial Protein Expression Data Yuju Lee Edward L Y Chen Darren C H Chan Anuroopa Dinesh Somaieh Afiunizadeh Conor Klamann Alina Selega Miralem Mrkonjic Hartland W Jackson Kieran R Campbell by Yuju Lee & Edward L. Y. Chen & Darren C. H. Chan & Anuroopa Dinesh & Somaieh Afiuni-zadeh & Conor Klamann & Alina Selega & Miralem Mrkonjic & Hartland W. Jackson & Kieran R. Campbell instant download after payment.

Spatial protein expression technologies can map cellular content and organization by simultaneously quantifying the expression of >40 proteins at subcellular resolution within intact tissue sections and cell lines. However,necessary image segmentation to single cells is challenging and error prone,easily confounding the interpretation of cellular phenotypes and cell clusters.To address these limitations, we present STARLING, a probabilistic machinelearning model designed to quantify cell populations from spatial proteinexpression data while accounting for segmentation errors. To evaluate performance, we develop a comprehensive benchmarking workflow by generating highly multiplexed imaging data of cell line pellet standards withcontrolled cell content and marker expression and additionally established ascore to quantify the biological plausibility of discovered cellular phenotypeson patient-derived tissue sections. Moreover, we generate spatial expressiondata of the human tonsil—a densely packed tissue prone to segmentationerrors—and demonstrate cellular states captured by STARLING identify knowncell types not visible with other methods and enable quantification of intraand inter- individual heterogeneity.